Daniel Voigt Godoy - Deep Learning with PyTorch Step-by-Step A Beginner’s Guide-leanpub

• is_valid_file: A function that checks if a file is corrupted or not.
Let’s create a dataset then:
Temporary Dataset
1 temp_transform = Compose([Resize(28), ToTensor()])
2 temp_dataset = ImageFolder(root='rps', transform=temp_transform)
We’re using only the transform optional argument here, and keeping
transformations to a minimum. First, images are resized to 28x28 pixels (and
automatically transformed to the RGB color model by the PIL loader, thus losing
the alpha channel), and then are converted to PyTorch tensors. Smaller images will
make our models faster to train, and more "CPU-friendly." Let’s take the first image
of the dataset and check its shape and corresponding label:
temp_dataset[0][0].shape, temp_dataset[0][1]
Output
(torch.Size([3, 28, 28]), 0)
Perfect!
"Wait, where is the standardization you promised?"
Standardization
To standardize data points, we need to learn their mean and standard deviation
first. What’s the mean pixel value of our rock paper scissors images? And standard
deviation? To compute these, we need to load the data. The good thing is, we have
a (temporary) dataset with the resized images already! We’re only missing a data
loader.
Temporary DataLoader
1 temp_loader = DataLoader(temp_dataset, batch_size=16)
No need to bother with shuffling, as this is not the data loader we’ll use to train the
420 | Chapter 6: Rock, Paper, Scissors